Traffic Pattern Recognition Method with XGBoost Based on Multi-scale Features

Traffic pattern recognition belongs to a branch of scene recognition and has become a hot research field. Correctly identifying the transportation mode used by users to travel plays a vital role in promoting the development of situational recognition. In the field of traffic pattern recognition research, many recognition methods use GPS, the combination of GPS and WiFi, and the combination of GSM and WiFi to obtain data, and use LR, SVM and deep learning models to make predictions. However, these methods have some disadvantages, such as poor WiFi signal in some outdoor environments, resulting in failure to obtain user data, GPS being interfered by the external environment, resulting in inaccurate data acquisition and other issues, and the models they use have problems such as low prediction accuracy or prolonged prediction consumption. In response to these problems, this paper uses multi-source sensors in mobile phones to obtain data, and preprocesses sensor data through statistics and signal processing methods to generate features at different scales, and finally uses XGBoost to identify various traffic modes. Extensive experiments are carried out on the method proposed in this paper, and the effectiveness of the proposed method is demonstrated by comparing with two state-of-the-art methods.